experiment(repeat, without preceding reverts) demonstrate that std-fs performs better because it hits the page cache

... by forcing each write system call to go to disk

test_bulk_insert[neon-release-pg14-tokio-epoll-uring].wal_written: 346 MB
test_bulk_insert[neon-release-pg14-tokio-epoll-uring].wal_recovery: 93.417 s

test_bulk_insert[neon-release-pg14-std-fs].wal_written: 346 MB
test_bulk_insert[neon-release-pg14-std-fs].wal_recovery: 86.009 s

=> ~8% instead of 2x difference
This commit is contained in:
Christian Schwarz
2024-03-13 15:03:24 +00:00
parent 578a2d5d5f
commit d66ccbae5e
2 changed files with 11 additions and 4 deletions

View File

@@ -200,7 +200,7 @@ def wait_for_last_record_lsn(
lsn: Lsn,
) -> Lsn:
"""waits for pageserver to catch up to a certain lsn, returns the last observed lsn."""
for i in range(100):
for i in range(10000):
current_lsn = last_record_lsn(pageserver_http, tenant, timeline)
if current_lsn >= lsn:
return current_lsn